Mark Hasegawa-Johnson received his Ph.D. from MIT in 1996. He is an assistant professor in the University of Illinois department of Electrical and Computer Engineering and a full-time faculty member in the Artificial Intelligence group at the Beckman Institute. His field of interest is speech production and recognition by humans and computers, including landmark-based speech recognition, integration of prosody in speech recognition and understanding, audiovisual speech recognition, computational auditory scene analysis, and biomedical imaging of the muscular and neurological correlates of speech production and perception.

Member, Articulograph International Steering Committee; CLSP Workshop leader, "Landmark-Based Speech Recognition" (2004), Invited paper; NAACL workshop on Linguistic and Higher-Level Knowledge Sources in Speech Recognition and Understanding (2004); List of faculty rated as excellent by their students (2003); NSF CAREER award (2002); NIH National Research Service Award (1998).

Human speech perception brings together abilities that have evolved either biologically or culturally over very long time periods in order to simultaneously extract semantic, phonemic, and paralinguistic information from a robust but complicated time-frequency code. Machine learning techniques excel at finding the optimum setting of parameters for a pre-specified speech recognition model structure, but machine learning techniques are not very good at choosing the right model structure. Dr. Hasegawa-Johnson's research seeks to apply higher-level knowledge from linguistics and psychology in order to specify the structure of machine learning models for automatic speech recognition. For example, machine learning models are capable of learning the class-conditional distributions of acoustic parameter vectors, but in speech recognition, it is not always clear how the "class" should be defined. The landmark-based speech recognition theory of Ken Stevens, based on several decades of linguistics research, suggests that phoneme boundaries form more acoustically invariant classes than do phoneme segments. Based on Stevens' theory, one of Dr. Hasegawa-Johnson's current research programs seeks to develop large vocabulary speech recognition algorithms using phoneme boundaries rather than phoneme segments as the fundamental phonological class. Likewise, several centuries of linguistic research clearly demonstrate that prosody (the melody and rhythm of natural language) influences the acoustic implementation of speech, but use of prosody in automatic speech recognition has been difficult because of the vast number of variables that have been proposed to bear salient information. By collaborating with University of Illinois linguists (Jennifer Cole and Chilin Shih), Dr. Hasegawa-Johnson has been able to select two binary prosodic distinctions considered by linguists to have the most dramatic acoustic and syntactic impact, and to show that explicit encoding of these prosodic distinctions into an automatic speech recognizer leads to reduced word error rate.